w2v2-queyu_w2v2Phoneme

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 1.8733
• Wer: 0.6796
• Cer: 0.2622

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2000
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
6.714	4.7619	100	4.2833	1.0	1.0
3.5176	9.5238	200	3.4836	1.0	1.0
3.454	14.2857	300	3.4308	1.0	1.0
3.382	19.0476	400	3.2659	1.0	0.9578
2.7759	23.8095	500	2.1766	0.9963	0.6285
1.5864	28.5714	600	1.4470	0.9214	0.3538
1.0146	33.3333	700	1.2868	0.7693	0.2964
0.7289	38.0952	800	1.2224	0.7606	0.2701
0.5847	42.8571	900	1.4075	0.7207	0.2762
0.444	47.6190	1000	1.3511	0.6733	0.2482
0.3747	52.3810	1100	1.5480	0.7057	0.2808
0.3487	57.1429	1200	1.6366	0.7207	0.2786
0.3123	61.9048	1300	1.7258	0.6995	0.2652
0.2766	66.6667	1400	1.6491	0.7244	0.2725
0.3014	71.4286	1500	1.6718	0.6920	0.2753
0.25	76.1905	1600	1.7745	0.7369	0.2786
0.2651	80.9524	1700	1.7571	0.7170	0.3015
0.2502	85.7143	1800	1.7912	0.7656	0.2993
0.252	90.4762	1900	1.8021	0.7868	0.3131
0.2402	95.2381	2000	1.6950	0.7182	0.3062
0.1208	100.0	2100	1.8733	0.6796	0.2622

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 3.0.0
• Tokenizers 0.22.1